Add Ornith-1.0-9B experimental slug (ik-llama/ornith9b-single) (#477)
DeepReinforce agentic-coding RL fine-tune — Qwen3-Next dense-FFN HYBRID (arch=qwen35: 8 full-attn + 24 GDN/DeltaNet layers, NON-MoE), single 3090, Q4_K_M + q8_0 KV, full 262K (KV only 4.25 GiB — just 8/32 layers carry full-attn GQA KV; 13.4 GiB total). Drafter-free ngram self-spec on ik_llama (works despite the DeltaNet hybrid). Full gate PASS: bench ~102 TPS, verify-stress 8/8, soak PASS, 8-pack 91/150 off / 95 on. 🧪 niche only — gemma-4-12b beats it on quality (105) + speed; pick for the lean 13.4 GiB footprint. Catalog suite 58/58 green.
Claude-Session: https://claude.ai/code/session_01EfF565T9eSLaqGzidyJ1Pm
Co-authored-by: noonghunna <[email protected]>
Co-authored-by: Claude Opus 4.8 <[email protected]>
